The X20's Green Problem

Started 11 months ago | Discussions thread
Nukunukoo
Regular MemberPosts: 353Gear list
Like?
Re: Non-issue - it's simply 'X-Trans' aliasing (no 'anti-aliasing' filter)...
In reply to Mark H, 11 months ago

Mark H wrote:

steras wrote:

which I managed to reproduce. However, when I tilted the camera 30-40 degrees relative to the target the vertical pattern in the green target disappeared. Any thoughts?

This test is most probably just revealing that the 'X-Trans' sensor is still susceptible to aliasing in the extreme circumstance that this particular test presents...

Why? Well, quite simply because there is no 'anti-aliasing filter, and the X-Trans CFA pattern still has a regular repeating pattern despite being very slightly less regular than a Bayer pattern.

The test targets are not a continuous spirals (whether green, red, or blue) - the coloured spiral targets are comprised of separated columns of pixels with gaps between the columns.

The pixels of the display test target, are simply 'aliasing' with the pixels of the camera sensor.

The aliasing reduced/disappeared when the camera was tilted 30-40 degrees, because the pixels of the display are no longer quite so closely aligned with those of the camera sensor.

Repeat the same test but with a continuous colour tone printed target (anyone got a nice laser printer handy?) and I'm pretty sure you won't get the same effect at all.

Similarly, defocus slightly, and I would predict the problem will be greatly reduced, if not disappear.

The problem really isn't with the camera at all - it's just a problem you might expect to encounter if you photograph a display made of 'pixels' with a camera sensor made of 'pixels' - especially in the absence of an 'anti-aliasing' (low pass) filter.

The reason the aliasing pattern differs between the green and that of the red and blue, is explained by the different pattern arrangement of the sensor's Gn pixels compared to the Rd and Be patterns, and also because the green channel has a higher resolution capability by virtue of it's greater pixel density/count - the pattern produced is a function of the CFA pattern and the demosaic interpolation algorithms.

To repeat - this is probably not really a camera/sensor problem at all - it's actually a slightly flawed test scenario.

A not-issue for 99.9% of photography - unless your favourite photographic is photographing RGB pixel displays.

[Edit: I see the later thread message with the printed target , pretty much supports my prediction above.]

Two Things:

(1) The distance between pixels on the display is less than 3 pels between strokes on the captured images. That's smaller than a single X-Trans block.

(2) The printed image sample shows distances between 8-11 pixels, higher than the X-Trans 6-pixel size at capture. To create the aliasing and demosaicing phenomenon, the distance and stroke width must be smaller than 6 pixels. Also, during resizing of any crop, the nearest neighbor method should be used to remove any edge smoothing. In addition, printouts "contaminate" the color test especially on CMYK-only units for obvious reasons.

Never said the test was flawless. It was crudely done but it does show how the X-Trans translate Green differently. And on that subject, I'm curious on what kind of result will I get if I did a black background (instead of white) on a totally dark room. Remember, AMOLED pels are transmissive, not transluscent like LCDs and LED backlits.

But the experiment was more to satisfy my curiosity on how DPR (and I) noticed some situations were greens were more smudged than normal. You are right though, such "flaws" only satisfy the fetish of the pixel-peepers!

Liz.

 Nukunukoo's gear list:Nukunukoo's gear list
Fujifilm X20 Nikon D300S Nikon D600 Nikon D7100 +1 more
Reply   Reply with quote   Complain
Post (hide subjects)Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark post MMy threads
Color scheme? Blue / Yellow